Hi, and thanks for releasing MapEx. While porting the probabilistic information gain to a different map predictor, I believe I found a mismatch between the paper and (get_free_points) in scripts/simple_mask_utils.py.
Paper (§IV-C): each ray starts with Δ = 0 and accumulates the mean-map occupancy of every pixel it traverses; the ray stops once the running sum Δ ≥ ε (with ε = 0.8). PIPE Planner, which reuses this raycast, states the same.
Code: In get_free_points accum_hit_prob = 0 is reset at the top of the inner (per-cell) loop.
Is the per-pixel reset intentional, or should Δ persist across the ray as described in IV-C? Happy to open a PR if a fix is wanted. Thanks!
Hi, and thanks for releasing MapEx. While porting the probabilistic information gain to a different map predictor, I believe I found a mismatch between the paper and (get_free_points) in
scripts/simple_mask_utils.py.Paper (§IV-C): each ray starts with Δ = 0 and accumulates the mean-map occupancy of every pixel it traverses; the ray stops once the running sum Δ ≥ ε (with ε = 0.8). PIPE Planner, which reuses this raycast, states the same.
Code:
In get_free_pointsaccum_hit_prob = 0 is reset at the top of the inner (per-cell) loop.Is the per-pixel reset intentional, or should Δ persist across the ray as described in IV-C? Happy to open a PR if a fix is wanted. Thanks!